Long-term optimized operation has been looking to unit commitment to replace the traditional power flow for more accurate modeling of system operation. Alongside the intuitive advantage of representing not only the production of the generation units of consecutive time periods but also their commitment status, comes the heavier computational burden for the multi-unit long-term model. In fact, representing the commitment status of each unit with a binary variable in the corresponding constraints alone complicates the optimization, let alone other accessory variables, such as those for switch actions. Moreover, unit commitment is usually conducted over a narrow window of time, e.g., one day or one week with an hourly resolution; ordinary unit commitment models that run over an annual load profile will be hindered from even converging to a solution by heavy computation. As such, clustering techniques are proposed to be equipped with the unit commitment in two dimensions, one to select representative periods for the load profile of a long horizon, the other to group homogeneous or similar units. As the former is conducted in the time dimension, it can be named the temporal clustering, and the latter hence named the spatial clustering. As a result, this new method with both is therefore named the tempo-spatially clustered unit commitment. The case study on a 39-node 17-unit system proves the efficacy and efficiency of the proposed unit commitment approach.
Keywords clustering, machine learning, unit clustering, unit commitment, representative days